john l. casti the computer as a laboratorycnls.lanl.gov/~toro/agents/bizsim.pdf · 2014-09-24 ·...
TRANSCRIPT
BizSim
The World of Business—in a Box
John L. Casti
The Computer as a Laboratory
The central process distinguishing science from its competitors—religion, music, litera-
ture, mysticism—in the reality-generation business is the so-called scientific method. An
integral part of this method by which we arrive at scientific “truth,” is the ability to do
controlled, repeatable laboratory experiments by which hypotheses about the phenomenon
under investigation can be tested. It is just such experiments that on a good day lead to
the theories and paradigms constituting today’s “scientific” world view. And, more than
anything else, it is the inability to perform experiments of this type that separate the
natural sciences from the worlds of social and behavioral phenomena. In the latter, we
have no way of doing the experiments necessary to create a bona fide scientific theory of
processes like stock market dynamics, road-traffic flow, and organizational restructuring.
In an earlier, less discerning era, it was often claimed that the realm of human social
behavior was beyond the bounds of scientific analysis, simply because human beings are
“complex”, “unpredictable”, “display free will”, “act randomly”, and so on and so forth.
It’s hard to believe that any modern system theorist would do anything but laugh at such
childish and naıve attitudes to the creation of workable and worthwhile scientific theories
of social and behavioral phenomena. The major barrier to bringing the social beneath
the umbrella of science is not the non-explanations just given in quotes, but the fact that
until now we have had no way to test hypotheses and, therefore, make use of the scientific
method in the creation of theories of social behavior. Now we do. And that laboratory in
which we do our experiments is the digital computer. Let me illustrate with an example
from the world of finance.
Booms and Busts, Bubbles and Crashes
In the fall of 1987, W. Brian Arthur, an economist from Stanford, and John Holland, a
computer scientist from the University of Michigan, were sharing a house in Santa Fe while
both were visiting the Santa Fe Institute. During endless hours of evening conversations
over numerous beers, Arthur and Holland hit upon the idea of creating an artificial stock
market inside a computer, one that could be used to answer a number of questions that
people in finance had wondered and worried about for decades. Among those questions
were:
• Does the average price of a stock settle down to its so-called fundamental value—the
value determined by the discounted stream of dividends that one can expect to receive
by holding the stock indefinitely?
• Is it possible to concoct technical trading schemes that systematically turn a profit
greater than a simple buy-and-hold strategy?
• Does the market eventually settle into a fixed pattern of buying and selling? In other
words, does it reach “stationarity”?
• Alternately, does a rich “ecology” of trading rules and price movements emerge in the
market?
Arthur and Holland knew that the conventional wisdom of finance argued that today’s
price of a stock is simply the discounted expectation of tomorrow’s price plus the dividend,
given the information available about the stock today. This theoretical price-setting proce-
dure is based on the assumption that there is an objective way to use today’s information
to form this expectation. But the information available typically consists of past prices,
trading volumes, economic indicators, and the like. So there may be many perfectly defen-
sible ways based on many different assumptions to statistically process this information in
order to forecast tomorrow’s price. For example, we could say that tomorrow’s price will
equal today’s price. Or we might predict that the new price will be today’s price divided
by the dividend rate. And so on and so forth.
2
The simple observation that there is no single, best way to process information led
Arthur and Holland to the not-very-surprising conclusion that deductive methods for fore-
casting prices are, at best, an academic fiction. As soon as you admit the possibility
that not all traders in the market arrive at their forecasts in the same way, the deductive
approach of classical finance theory, which relies upon following a fixed set of rules to
determine tomorrow’s price, begins to break down. So a trader must make assumptions
about how other investors form expectations and how they behave. He or she must try to
psyche out the market. But this leads to a world of subjective beliefs and to beliefs about
those beliefs. In short, it leads to a world of induction in which we generalize rules from
specific observations rather than one of deduction.
In order to address these kinds of questions, Arthur, Holland and their colleagues
constructed an electronic stock market, in which they could manipulating trader’s strate-
gies, market parameters, and all the other things that cannot be done with real exchanges.
The traders in this market are assumed to each summarize recent market activity by a
collection of descriptors, which involve verbal characterization like “the price has gone up
every day for the past week,” or “the price is higher than the fundamental value,” or “the
trading volume is high.” Let us label these descriptors A, B, C, and so on. In terms of the
descriptors, the traders decide whether to buy or sell by rules of the form: “If the market
fulfills conditions A, B, and C, then buy, but if conditions D, G, S, and K are fulfilled,
then hold.” Each trader has a collection of such rules, and acts in accordance with only
one rule at any given time period. This rule is the one that the trader views as his or her
currently most accurate rule.
As buying and selling goes on in the market, the traders can reevaluate their different
rules by assigning higher probability of triggering a given rule that has proved profitable
in the past, and/or by recombining successful rules to form new ones that can then be
tested in the market. This latter process is carried out by use of what is called a genetic
algorithm, which mimics the way nature combines the genetic pattern of males and females
of a species to form a new genome that is a combination of those from the two parents.
A run of such a simulation involves initially assigning sets of predictors to the traders
at random, and then beginning the simulation with a particular history of stock prices,
3
interest rates, and dividends. The traders then randomly choose one of their rules and
use it to start the buying-and-selling process. As a result of what happens on the first
round of trading, the traders modify their estimate of the goodness of their collection of
rules, generate new rules (possibly), and then choose the best rule for the next round of
trading. And so the process goes, period after period, buying, selling, placing money in
bonds, modifying and generating rules, estimating how good the rules are, and, in general,
acting in the same way that traders act in real financial markets.
A typical frozen moment in this artificial market is displayed in Figure 1. Moving
clockwise from the upper left, the first window shows the time history of the stock price
and dividend, where the current price of the stock is the black line and the top of the
grey region is the current fundamental value. The region where the black line is much
greater than the height of the grey region represents a price bubble, whereas the market
has crashed in the region where the black line sinks far below the grey. The upper right
window is the current relative wealth of the various traders, and the lower right window
displays their current level of stock holdings. The lower left window shows the trading
volume, where grey is the number of shares offered for sale and black is the number of
shares that traders have offered to buy. The total number of trades possible is then the
smaller of these two quantities, because for every share purchased there must be one share
available for sale.
After many time periods of trading and modification of the traders’ decision rules,
what emerges is a kind of ecology of predictors, with different traders employing differ-
ent rules to make their decisions. Furthermore, it is observed that the stock price always
settles down to a random fluctuation about its fundamental value. However, within these
fluctuations a very rich behavior is seen: price bubbles and crashes, market moods, over-
reactions to price movements, and all the other things associated with speculative markets
in the real world.
The agents in the stockmarket simulation are individual traders. A quite different type
of business simulation emerges when we want to look at an entire industry, in which case
the agents become the individual firms constituting that industry. The world’s catastrophe
insurance industry served as the focus for just such a simulation exercise called Insurance
4
Figure 1. A frozen moment in the surrogate stock market.
World, carried out by the author and colleagues at the Santa Fe Institute and Intelligize,
Inc. over the past couple of years.
Insurance World
As a crude, first-cut, the insurance industry can be regarded as an interplay among three
components: firms, which offer insurance, clients, who buy it, and events, which determine
the outcomes of the “bets” that have been placed between the insurers and their clients.
In Insurance World, the agents consist of primary casualty insurers and the reinsurers,
the firms that insure the insurers, so to speak. The events are natural hazards, such as
hurricanes and earthquakes, as well as various external factors like government regulators
and the global capital markets.
Insurance World is a laboratory for studying questions of the following sort:
5
• Optimal Uncertainty: While insurers and reinsurers talk about getting a better
handle on uncertainty so as to more accurately assess their risk and more profitably price
their product, it’s self-evident that perfect foreknowledge of natural hazards would spell
the end of the insurance industry. On the other hand, complete ignorance of hazards is
also pretty bad news, since it means there is no way to weight the bets the firms make
and price their product. This simple observation suggests that there is some optimal level
of uncertainty at which the insurance—but perhaps not their clients—can operate in the
most profitable and efficient fashion. What is that level? Does it vary across firms? Does
it vary between reinsurers, primary insurers, and/or end consumers?
• Industry Structure: In terms of the standard metaphors used to characterize organ-
izations—a machine, a brain, an organism, a culture, a political system, a psychic prison—
which type(s) most accurately represents the insurance industry? And how is this picture
of the organization shaped by the specific “routines” used by the decisionmakers in the
various components making up the organization?
The simulator calls for the management of each firm to set a variety of parameters
having to do with their desired market share in certain regions for different types of hazards
and level of risk they want to take on, as well as to provide a picture of the external
economic climate (interest rates, likelihood of hurricanes/earthquakes, inflation rates and
so forth). The simulation then runs for 10 years in steps of one quarter, at which time
a variety of outputs can be examined. For instance, Figure 2 shows the market share for
Gulf Coast hurricane insurance of the five primary insurers in this toy world, under the
assumption that the initial market shares were almost identical—but not quite. In this
experiment, firm 2 has a little larger initial market share than any of the other firms, a
differential advantange that it then uses to squeeze out all the other firms at the end of
the ten-year period. This is due to the “brand effect,” in which buyers tend to purchase
insurance from companies that they know about.
As a final example of what simulation and business have to say to each other, con-
sider the movement of shoppers in a typical supermarket. This world is dubbed SimStore
by Ugur Bilge of SimWorld, Ltd. and Mark Venables at J. Sainsbury in London, who
collaborated with the author on its creation.
6
Figure 2. Market share distribution for five primary insurers.
SimStore
The starting point for SimStore is a real supermarket in the Sainsbury chain, one located
in the London region of South Ruislip. The agents are individual shoppers who frequent
this store. These electronic shoppers are dropped into the store, and then make their way
to the various locations in the store by rules such as “wherever you are now, go to the
location of the nearest item on your shopping list,” so as to gather all the items they want
to purchase.
As an example of one of the types of outputs generated by SimStore, customer checkout
data are used to calculate customer densities at each location. Color codes are with
descending order: blue, red, purple, orange, pink, green, cyan, grey and nothing. Using
the Manhattan metric pattern of movement, in which a customer can only move along the
aisles of the store, all locations above 30 percent of customer densities have been linked
to form a most popular customer path. Once this path is formed a genetic algorithm will
minimize (or maximize!) the length of the overall shopping path.
7
In the same store, this time each individual customer path has been internally cal-
culated using the simple “nearest neighbor” rule noted above. All customer paths have
been summed for each aisle, in order to calculate the customer path densities. These den-
sities are displayed in Figure 3 as a relative density map using the same color code just
mentioned.
Figure 3. Customer densities along each aisle in the simulated store.
Simulation is Good for Business
Large-scale, agent-based simulations of the type discussed here are in their infancy. But
even the preliminary exercises outlined here show the promise of using modern computing
technology to provide the basis for doing experiments that have never been possible before.
Even better, these experiments are exactly the sort called for by the scientific method—
controlled and repeatable—so that for the first time in history we have the opportunity to
actually create a science of human affairs. If I were placing bets on the matter, I’d guess
8
that the world of business and commerce will lead the charge into this new science that
will form during the 21st century.
References
Casti, J. Would-Be Worlds. New York: Wiley, 1997.
9